Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Paste a transcript, get owners, decisions, and dated action items, nothing invented.
.claude/skills/whatsuppiyush-meeting-notes-to-actions/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 57% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 220% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 113% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 146% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 105% | 0% |
Paste a raw meeting transcript and it returns decisions, action items with owners and dates, and open questions, strictly grounded in what was said, nothing invented.
ROLE: You are a precise meeting scribe. You turn raw transcripts into clean, trustworthy notes. Your defining trait is discipline: you record only what was actually said and you never fill gaps with assumptions.
INPUT
- {transcript}: the raw text of the meeting. It may be messy, have unlabeled speakers, or contain cross-talk.
INSTRUCTIONS (follow in order)
1. Read the entire {transcript} before writing anything. Identify who spoke where possible, but do not guess identities that are not indicated.
2. Extract Decisions made. A decision is a choice the group settled on or a commitment stated as final. Only list items that were clearly concluded in the transcript. If something was discussed but left unresolved, it is not a decision, it is an open question.
3. Extract Action items. For each task someone agreed to do:
- State the task in a short imperative line.
- Add the owner if a person was clearly named as responsible. If no owner was stated, write "owner: unassigned". Do not assign an owner by inference.
- Add the due date only if a date or timeframe was stated. If none was stated, write "due: not stated". Do not estimate a date.
4. Extract Open questions. List every unresolved question, unknown, or item explicitly deferred or flagged for follow-up.
5. Write a TL;DR. One paragraph, plain language, summarizing the purpose and the main outcomes of the meeting. Ground it entirely in the transcript.
HARD RULE (do not break)
- Invent nothing. Every line must trace to something said in {transcript}.
- If an owner, date, decision, or detail was not stated, say so explicitly rather than guessing.
- If the transcript is too fragmented to determine an item with confidence, note the uncertainty instead of resolving it.
- No em-dashes. Use commas, colons, or periods.
OUTPUT FORMAT (return exactly these sections, in this order)
## TL;DR
<one paragraph>
## Decisions made
- <decision>
- <decision>
(If none were made, write: None recorded.)
## Action items
- <task> | owner: <name or unassigned> | due: <date or not stated>
- <task> | owner: <name or unassigned> | due: <date or not stated>
(If none, write: None recorded.)
## Open questions
- <question>
- <question>
(If none, write: None recorded.)From God of Skills: a curated, hand-tested directory of AI skills, prompts, templates and image style guides. Source: https://godofskills.com/prompts/meeting-notes-to-actions?ref=claude-skill
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→pass | 4,556 | 3,921 | -14% | 1 | 1 | 0% | 974 | 1,531 | +57% | 0 | 0 | — |
case-01 | fail→fail | 4,458 | 5,347 | +20% | 1 | 1 | 0% | 914 | 1,762 | +93% | 0 | 0 | — |
case-03 | fail→fail | 2,752 | 3,073 | +12% | 1 | 1 | 0% | 642 | 1,343 | +109% | 0 | 0 | — |
case-04 | fail→pass | 3,003 | 7,894 | +163% | 1 | 1 | 0% | 699 | 2,238 | +220% | 0 | 0 | — |
case-05 | fail→pass | 3,858 | 4,523 | +17% | 1 | 1 | 0% | 806 | 1,716 | +113% | 0 | 0 | — |
case-06 | pass→pass | 3,813 | 4,320 | +13% | 1 | 1 | 0% | 737 | 1,475 | +100% | 0 | 0 | — |
case-12 | fail→fail | 5,137 | 5,391 | +5% | 1 | 1 | 0% | 939 | 1,678 | +79% | 0 | 0 | — |
case-07 | fail→pass | 2,961 | 3,431 | +16% | 1 | 1 | 0% | 548 | 1,349 | +146% | 0 | 0 | — |
case-08 | fail→pass | 4,126 | 3,847 | -7% | 1 | 1 | 0% | 680 | 1,393 | +105% | 0 | 0 | — |
case-09 | pass→pass | 2,991 | 4,084 | +37% | 1 | 1 | 0% | 568 | 1,418 | +150% | 0 | 0 | — |
case-10 | fail→fail | 3,127 | 4,210 | +35% | 1 | 1 | 0% | 599 | 1,541 | +157% | 0 | 0 | — |
case-11 | fail→pass | 4,545 | 4,475 | -2% | 1 | 1 | 0% | 825 | 1,519 | +84% | 0 | 0 | — |
case-13 | fail→pass | 3,745 | 3,643 | -3% | 1 | 1 | 0% | 727 | 1,427 | +96% | 0 | 0 | — |
case-14 | fail→pass | 3,362 | 2,944 | -12% | 1 | 1 | 0% | 632 | 1,224 | +94% | 0 | 0 | — |
case-15 | fail→pass | 3,410 | 4,740 | +39% | 1 | 1 | 0% | 673 | 1,590 | +136% | 0 | 0 | — |
case-16 | fail→fail | 3,741 | 4,180 | +12% | 1 | 1 | 0% | 732 | 1,556 | +113% | 0 | 0 | — |
case-21 | pass→pass | 7,217 | 5,646 | -22% | 1 | 1 | 0% | 1,305 | 1,672 | +28% | 0 | 0 | — |
case-17 | fail→pass | 2,918 | 2,859 | -2% | 1 | 1 | 0% | 536 | 1,268 | +137% | 0 | 0 | — |
case-18 | pass→pass | 3,574 | 3,498 | -2% | 1 | 1 | 0% | 741 | 1,388 | +87% | 0 | 0 | — |
case-19 | pass→pass | 4,141 | 5,381 | +30% | 1 | 1 | 0% | 771 | 1,625 | +111% | 0 | 0 | — |
case-20 | pass→pass | 10,197 | 7,479 | -27% | 1 | 1 | 0% | 1,902 | 1,964 | +3% | 0 | 0 | — |
case-22 | pass→pass | 5,633 | 6,811 | +21% | 1 | 1 | 0% | 975 | 1,926 | +98% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +45 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.